Cartesia
Low-latency AI voices (Sonic-3.6) for live agents, dubbing and fast VO drafts
Cartesia makes the Sonic text-to-speech models, the Ink speech-to-text models and a voice-agent platform built on them. Sonic-3.6 went generally available on Aug 27, 2026 with 44 languages, 61 locales and a claimed response time under 90 ms. Aimed at developers building voice agents and real-time speech products rather than narration studios.
- 1 creator using
- 3 news items tracked
- 2024 first launched
- Freemium pricing
- Moderate learning curve
What Cartesia can do
- Infinite canvas No
- AI agent Yes
- Node / flow builder Not confirmed
- Public API Yes
- MCP server Yes
9 of 26 capabilities confirmed
| Capability | Supported | Details |
|---|---|---|
| Text to image | No | |
| Image editing | No | |
| Text to video | No | |
| Image to video | No | |
| Video to video | No | |
| Native audio in video | No | |
| Lip-sync | No | |
| Voice / TTS | Yes | Sonic-3.6 TTS, instant voice cloning from Pro, professional voice clones from Startup |
| Music & SFX | No | |
| Upscaling | No | |
| Character consistency | Partial | Saved voices and professional voice clones give a consistent voice across generations, not visual identity |
| Camera control | No | |
| Custom training | Partial | Professional voice clones trained on a speaker sample; no LoRA or custom model training |
| Third-party models | No |
| Capability | Supported | Details |
|---|---|---|
| Infinite canvas | No | |
| Node / flow builder | Not confirmed | |
| AI agent | Yes | Managed Agents voice-agent platform, billed at $0.06 per call-minute on paid plans |
| Chat assistant | Not confirmed | |
| Team collaboration | Not confirmed |
| Capability | Supported | Details |
|---|---|---|
| Public API | Yes | REST and WebSocket API, Cartesia-Version header required |
| MCP server | Yes | Official cartesia-mcp server with text_to_speech and voice tools |
| Mobile app | Not confirmed | |
| Desktop / local | Partial | On-prem and on-device deployment offered, through sales for enterprise |
| Open source | No |
| Capability | Supported | Details |
|---|---|---|
| Free tier | Yes | 20K credits per month |
| Commercial use | Yes | Commercial use license from the Pro plan up |
Yes Partial No Not confirmed by the vendor
What makes Cartesia different
- 01
Vendor-claimed response time under 90 ms on Sonic-3.6, the main reason to pick it over narration-first TTS tools
- 02
Same vendor covers TTS (Sonic), STT (Ink) and a voice-agent layer (Managed Agents), so one API key runs a full voice loop
- 03
Free tier of 20K credits per month (about 27 min of Sonic TTS) with no card needed to start testing
- 04
Official cartesia-mcp server exposes a text_to_speech tool to AI assistants
- 05
Deployable in cloud, on-prem or on-device per the homepage; on-prem and VPC go through sales
Latest from Cartesia
- 2w ago Cartesia Powers Multilingual Voice Interfaces Across Indian Financial Apps Cartesia partners with Raise Financial Services to deliver low-latency, multilingual text-to-speech across Indian retail investment platforms.
- Jun 30 Sonic Framework Standardizes Voice AI Performance Evaluation Cartesia released a practical framework for evaluating text-to-speech and speech-to-text models based on real-world latency and reliability rather than lab benchmarks.
- Sep 24 Cartesia Achieves GDPR Compliance for Enterprise Audio Workflows Cartesia now meets European data protection standards, allowing creators and studios to process voice synthesis with improved privacy security.
About Cartesia
Cartesia's product line in October 2026 is Sonic (text-to-speech), Ink (streaming speech-to-text) and Managed Agents, the voice-agent layer that ties them together. Sonic-3.5 went live in April 2026 and Sonic-3.6 followed in late August with more locales, more than 500 preset voices, Hinglish code-switching and faster generation. Ink-2, launched in May 2026, adds built-in turn detection and keyterm prompting for names and jargon, and is English-only for now. In September 2026 the library gained over 50 voices that speak up to 25 languages, and custom voice clones can be made multilingual.
For filmmakers the product is an API-first tool. There is a playground for auditioning voices, but no timeline or editor. Use it for rough-cut voice-over, multilingual test versions of a spot, interactive character dialogue, or the voice layer of a phone or kiosk installation. It is not a replacement for a recorded session when a performance needs range, and the vendor itself positions it for speed over long-form emotional nuance.
The pricing model changed in 2026 to monthly subscriptions measured in credits, with a free tier of 20K credits, which covers roughly 27 minutes of Sonic speech. Commercial-use rights start at the Pro plan at $5 per month.
What's new in Cartesia
-
Multilingual library voices
Over 50 library voices now speak up to 25 languages, and custom voice clones can be made multilingual.
Source: releasebot.io ↗ -
Sonic-3.6 released (GA)
44 languages, 61 locales, 500+ preset voices, Odia and Urdu added, Hinglish code-switching, response time under 90 ms as claimed by Cartesia.
Source: plushcap.com ↗ -
Ink-2 keyterm prompting and turn-detection controls
Up to 100 keyterms can be passed to improve recognition of names and terms, and turn detection can be tuned for latency or endpointing accuracy.
Source: releasebot.io ↗ -
Sonic-2 and Sonic-turbo sunset scheduled
Sonic-2, Sonic-turbo and the Sonic-3-2025-10-27 snapshot stop working after October 20, 2026, so API users need to move to Sonic-3.5 or 3.6.
Source: releasebot.io ↗ -
Professional voice clones on Sonic-3.5 and batch outbound calling
Professional voice clones became available on Sonic-3.5, custom sample rates from 8 to 44.8 kHz were added, and batch outbound calling supports up to 5,000 calls per request.
Source: releasebot.io ↗ -
Ink-2 streaming speech-to-text launched
Streaming STT with built-in turn detection, English only at launch. Cartesia's docs date the model May 22, 2026, while some trackers say July.
Source: releasebot.io ↗
Key Features
- Sonic-3.6 TTS: 44 languages, 61 locales, 500+ preset voices (GA Aug 2026)
- Vendor-claimed response time under 90 ms for streaming speech
- Instant voice cloning from Pro up; professional voice clones on Startup (2 slots) and Scale (4 slots)
- Multilingual library voices: 50+ voices speaking up to 25 languages (Sep 2026)
- Ink-2 streaming speech-to-text with built-in turn detection and keyterm prompting (up to 100 terms), English only
- Managed Agents voice-agent platform with telephony and batch outbound calling
- SSML, custom pronunciations, speed and volume controls, sample rates from 8 to 44.8 kHz
- REST and WebSocket API, plus official cartesia-mcp server with a text_to_speech tool
When to reach for it — and when to skip
Reach for it when…
- Lowest response-time claim in this catalog for speech generation (vendor figure, under 90 ms)
- Free tier with 20K credits a month, enough to audition voices and build a prototype
- TTS and STT from one vendor, so an agent loop needs one account and one API
- MCP server makes it usable from AI assistants without writing integration code
Skip it when…
- Built for speed; the vendor itself says less emotional nuance than ElevenLabs for narration
- No editor or timeline. Output is audio files or streams, so a filmmaker needs a DAW or NLE to place it
- Commercial use and instant voice cloning start at Pro; the free tier is for testing only
- Sonic-2 and Sonic-turbo stop working after Oct 20, 2026, so older integrations need a migration
Best For
✓ Ideal for
- Real-time voice agents for phone lines, kiosks and customer support
- Fast VO scratch tracks and rough-cut placeholders before a real session
- Multilingual test versions of a spot or trailer voice-over (44 languages)
- Interactive game NPC and installation dialogue where delay is noticeable
✗ Not built for
- Final narration or character performance where emotional range matters most (audition ElevenLabs first)
- Music, sound effects or video generation, which Cartesia does not offer
Working Tips from Filmmakers Using Cartesia
- 01 Audition voices in the playground on the Free tier first. 20K credits is roughly 27 minutes of Sonic speech, enough for a short spot's VO drafts
- 02 Pro at $5 per month is the cheapest plan with a commercial use license and instant voice cloning. Check the license before using a cloned voice in paid work
- 03 Put character and place names into Ink-2 keyterm prompting (up to 100 terms) when transcribing interviews, which cuts manual correction time
- 04 Set custom sample rates in the API (8 to 44.8 kHz) to match the delivery spec of your edit before importing into Premiere or Resolve
Cartesia pricing in 2026
| Plan | Price | What you get |
|---|---|---|
| Free | $0 monthly |
|
| Pro | $5 monthly |
|
| Startup | $49 monthly |
|
| Scale | $299 monthly |
|
| Enterprise | Custom custom |
|
Prices checked against Cartesia's pricing page on 9 Oct 2026. Vendors change plans often, so confirm before you buy.
Cartesia vs ElevenLabs
| Cartesia | ElevenLabs | |
|---|---|---|
| Price | Free (20K credits/mo) · Pro $5/mo · Startup $49/mo · Scale $299/mo · Enterprise custom | Free: 10k credits/mo. Starter from $5/mo billed annually ($6 monthly). Creator $18.33/mo annual ($22 monthly). Pro $99/mo. Scale $299/mo. Business $990/mo. |
| AI agent | Yes | Partial |
| Team collaboration | Not confirmed | Partial |
| Voice / TTS | Yes | Yes |
| Music & SFX | No | Yes |
| Character consistency | Partial | Partial |
| Custom training | Partial | Partial |
| Public API | Yes | Yes |
| MCP server | Yes | Not confirmed |
| Mobile app | Not confirmed | Yes |
| Desktop / local | Partial | Partial |
| Free tier | Yes | Yes |
| Commercial use | Yes | Yes |
| Free tier | 20K credits/mo (~27 min TTS) | 10K credits/mo |
| Lowest paid plan | Pro $5/mo, 100K credits | Starter $6/mo list ($1 first-month promo), 30K credits |
| Commercial use on paid plans | Pro and up | Starter and up |
| Latency claim | Under 90 ms (Sonic-3.6, vendor claim) | Not stated on pricing page |
Each column comes from that tool's own researched page. "Not confirmed" means the vendor doesn't say.
Open the full Cartesia vs ElevenLabs comparison →Use Cases
Tags
Sources
Facts on this page were checked against Cartesia's own pricing, product and changelog pages on , then cross-checked with independent coverage where vendor pages were silent. Anything we couldn't confirm is marked "Not confirmed" rather than guessed.
- Cartesia pricing page cartesia.ai
- Cartesia homepage cartesia.ai
- Cartesia release notes (Apr to Sep 2026, aggregator of Cartesia changelog) releasebot.io
- Introducing Sonic-3.6 plushcap.com
- Sonic-3.5 and Ink-2 voice agent benchmarks and Line pricing runtimewire.com
- Cartesia MCP server docs docs.cartesia.ai
- ElevenLabs pricing page (comparison) elevenlabs.io
Spotted something out of date? Tell us and we'll re-check it.
Discussion
No comments yet — be the first.